A French judge said the quiet part out loud this year. Carrefour, defending itself in court over an online grocery service that blind customers could not use, argued that its website met 71% of the applicable accessibility criteria. That is a serious number. It represents real work by real engineers over real sprints.

On 4 June 2026, the Tribunal judiciaire de Caen rejected it anyway. The court ordered Carrefour France to make carrefour.fr and its mobile app fully accessible within six months, with a penalty of 500 euros for every day it runs past the deadline. That deadline lands in early December. It was the first ruling under a national transposition of the European Accessibility Act to order a company to fix its site, and the first anywhere in the regime to name a mobile app explicitly. Deque has a good summary of the decision.

The logic is worth sitting with, because it dismantles the way most teams have been measuring this. A blind customer does not experience 71% compliance. They experience a checkout they cannot finish.

---

Where the numbers actually sit

The gap between what teams believe about their sites and what is measurably true has been widening, not closing.

• 95.9% of the top one million home pages fail basic automated WCAG checks.

  The WebAIM Million 2026 report found detectable failures on 95.9% of pages, up from 94.8% the year before. That reverses six consecutive years of slow improvement.



• The average page now carries 56.1 detected errors, up 10.1% in a year.

  Users with disabilities can expect a barrier on roughly one in every 26 page elements.



• Shopping sites are the worst performing category measured.

  Ecommerce home pages averaged 71 errors, 26.6% above the overall average. Pages built on Shopify averaged 75.1.



• Six failure types account for 96% of every error detected.

  They have been the same six for seven years running.



• US filings are on track for a record year.

  UsableNet counted more than 5,000 digital accessibility lawsuits in 2025 and projects roughly 6,176 for 2026 based on filings through mid June, with close to 79% aimed at ecommerce. Its monthly tracker logged 432 new filings in August alone.

Page complexity is the uncomfortable explanation. The average home page grew to 1,437 elements in February 2026, a 22.5% increase in twelve months. ARIA attributes hit 133 per page, up 27%, and pages using ARIA averaged 59.1 errors against 42 for pages without it. WebAIM's own conclusion points at heavier reliance on third-party frameworks and automated coding practices. We are shipping more markup per screen than ever, and a meaningful share of it is unlabelled.

---

1. Settle whether you are in scope, properly

The European Accessibility Act, Directive 2019/882, became enforceable on 28 June 2025. It applies based on where you sell, not where you are registered. A store in Sydney or Denver that ships to Dublin is inside it. The usual carve-out is for businesses under 10 employees and under 2 million euros in annual turnover, and the technical baseline is EN 301 549, which builds on WCAG 2.1 Level AA.

There is a transition window running to June 2030 for services already on the market before June 2025. Before you build a roadmap on that clause, note that carrefour.fr is considerably older than 2025 and the court gave it six months regardless. Treat the transition period as a cushion, not a plan.

2. Stop reporting a percentage

A conformance score is a claim about a sample, not a statement about whether a customer can buy something. It is also the exact defence that failed in Caen.

Replace the score with journey coverage. Pick your five revenue-critical paths: find a product, add to basket, register, pay, request a refund. A path either works end to end for a screen reader and keyboard user, or it does not. Five booleans tell your board more than a 71 ever will.

3. Clear the six failures that cause 96% of errors

This is the least glamorous and highest yield work available to you. Low contrast text appears on 83.9% of home pages, missing image alternative text on 53.1%, missing form input labels on 51%, empty links on 46.3%, empty buttons on 30.6%, and missing document language on 13.5%.

None of these require a redesign. Contrast is a token change in your design system. Document language is one attribute. Empty links and buttons are almost always icon controls that shipped without an accessible name. A focused sprint clears most of it.

4. Label every input in the checkout

One third of all form inputs on the pages WebAIM tested, 33.1%, were not properly labelled by any method. Home pages now carry 6.9 inputs on average, up 36% in three years.

Placeholder text is not a label. It disappears on focus and screen readers treat it inconsistently. Every field needs a real label, a programmatic association, an error message tied to the field rather than floating above the form, and a focus state a sighted keyboard user can actually see.

5. Unplug your mouse and try to buy something

Thirty minutes, no tooling, no budget. Tab through your own purchase flow. Watch for focus that disappears behind a sticky header, modals that let focus escape to the page underneath, carousels that trap you, and cookie banners that block interaction without ever receiving focus themselves.

Keyboard access is the foundation every assistive technology sits on. If the flow fails here, nothing built on top of it will hold.

6. Audit the code you did not write

Most accessibility debt arrives by dependency. WebAIM's data shows pages using popular carousel and lightbox libraries averaging 25% to 45% more errors than the sample average, reCAPTCHA adding 7.7 errors on average, and ad networks correlating strongly with higher error counts.

Inventory the third parties in your critical paths: the cookie consent layer, the payment iframe, the review widget, the chat launcher, the search overlay. Each one either meets your standard or gets replaced. Your vendor's defect becomes your legal exposure the moment it renders on your domain.

7. Understand what an overlay does and does not do

Accessibility widgets are not fraud. They genuinely help some users, and a contrast toggle or a text resizer is a real convenience for someone who needs it. That concession matters, because the industry argument against overlays is usually made badly.

Here is what they do not do. They do not repair the underlying markup, and they do not stop a claim. UsableNet's mid year data shows roughly one in five sued companies was running a widget at the time. In August 2026, 134 of 432 defendants had one installed. Sell an overlay as a convenience layer if you like. Do not sell it internally as a compliance solution.

8. Move accessibility into the pipeline

Carrefour argued in court that its site changes daily, which makes full real-time accessibility genuinely difficult. That is true, and it is not a defence. It is an argument for a different process.

Put contrast ratios into your design tokens so a failing colour cannot be specified. Run an automated checker in continuous integration and fail the build on new violations rather than reporting on them. Add an accessible-name check to your component library tests. Book manual testing with assistive technology users twice a year. An annual audit produces a document. A pipeline produces a site that stays fixed between audits.

---

The argument that does not fit in a compliance deck

Legal exposure is the least interesting reason to do this work, and it is the reason everyone leads with because it is the one that unlocks budget.

The more useful framing is that accessibility failures are precise, reproducible descriptions of your site breaking. An unlabelled input is a form that some customers cannot complete. An empty button is a control that announces nothing to anyone who is not looking directly at it. Low contrast text is copy your customer is squinting at on a phone in daylight. These are not compliance artefacts. They are conversion defects that happen to be illegal in a growing number of jurisdictions.

Enforcement across the EU is still early and uneven. No confirmed fine has been issued under a national EAA transposition as of mid 2026, and the Carrefour order is a forward-looking penalty rather than a sanction for past conduct. A parallel case against Auchan was dismissed in May on a revenue threshold, though the same court found the site failing 13 of 19 sections. Regulators in the Netherlands and Sweden have moved from information requests into active work, and German operators have been receiving private warning letters since 2025. The direction is not ambiguous, even if the pace still is.

Waiting to see how it settles is a position. It is just an expensive one to hold while your complexity keeps growing.

---

Three tests to run this week

• Run your five highest-traffic pages through WAVE.

  The free tool takes about ten minutes total and will surface most of the six common failures immediately.



• Complete one purchase with the keyboard only.

  Note every point where you lose the focus indicator or cannot reach a control.



• Turn off all images and read your product page.

  If the page becomes incomprehensible, your alternative text is doing nothing, and roughly one in four images on a typical home page has missing, questionable or repetitive alt text.

None of that costs anything. What it produces is a ranked list of real defects, which is a considerably better starting point than a percentage.

Need the audit, the remediation and the pipeline that keeps it fixed? Talk to Spark about making your site work for every customer you already paid to attract.